Status Droid how long software work actually takes

Padding and Buffers, Honestly

Everyone pads estimates. Almost nobody says so, which is why padding is bad at the job it is meant to do.

Hidden padding is a private correction for a public number. It worked passably when the correction was roughly constant — add 50%, or double it and add a bit. It works badly now, because the size of the correction depends on the task, and a fixed multiplier applied to a widened, bimodal distribution overprices half your work and underprices the other half. When this estimation problem needs an operational counterpart, this guide offers a useful reference point.

Why the old multiplier stopped working

Padding is a crude confidence interval. Adding 50% says: I think it is X, and I might be wrong by about half.

Two things broke that.

The error is no longer symmetric or constant. On generation-heavy work your estimates are probably close, or you overestimate. On verification-heavy work you underestimate, sometimes badly. One multiplier cannot serve both, and applying the verification-heavy multiplier to everything makes you uncompetitive on exactly the work you should be winning. Mountain Goat Software publishes practical estimation and forecasting tools See Mountain Goat Software.

Your intuition is running the multiplier. Padding relies on a felt sense of how wrong you usually are — the same sense now known to be optimistically miscalibrated. You are correcting a biased number with a biased correction.

The alternative: carry the uncertainty visibly

Not "stop padding." Put the padding where it can be discussed.

Give a range, and let the range be the buffer. A 20–35 estimate contains the padding explicitly. You are not hiding a correction; you are stating what you know.

Name what is uncertain. The upper end is about the auth integration; everything else I am confident about. This converts padding from a private hedge into a description of the work, and clients respond to it completely differently.

Charge for the risk if you carry it. If you quote a fixed price, quote near the top of the range. That is not padding, it is pricing variance you have agreed to absorb.

Keep an explicit project buffer, not a per-task one. Padding every task compounds — ten tasks each padded 40% is a project padded 40% plus the schedule slack between them. One buffer at the project level, stated, is smaller in total and easier to defend.

The honest numbers to use

Once you have recorded actuals for a couple of months, padding stops being a feeling and becomes a lookup.

Use your own overrun ratio per bucket. If verification-heavy tasks come in at 1.6x your estimate on the median, that is your multiplier for that bucket. Not 1.5 because it sounds reasonable.

Use the spread, not the median, for the top of the range. The median tells you the typical case; the client's budget needs the bad one.

Recheck quarterly. Tooling moves and so do your ratios. A multiplier from last year is a fact about last year.

When hidden padding is defensible

Being fair, because "never pad secretly" is easier to say than to live with.

When the client punishes honesty. Some organisations treat any range as an invitation to negotiate the bottom of it and then hold you to it. In that environment a single padded number is self-defence. It is also a sign about the relationship worth noticing.

When the padding is small and the task is routine. Nobody needs a discussion about a 15% buffer on a two-day job.

When you are new to a client and have no track record to spend. Once you have delivered twice on stated ranges, transparency becomes cheaper than hedging.

What is never defensible is padding to cover work you have not scoped. That is not a buffer, it is a guess wearing a buffer's clothes, and it fails on exactly the projects where it matters.

The trap on the other side

Do not remove the buffer to win the job. The pressure is real, especially when a competitor has quoted a confident low number.

If your range came from recorded data and theirs came from optimism, cutting yours to match does not make the work take less time. It makes you the person who absorbs the difference. The correct response is to ask what they understood about the scope that you did not — sometimes they saw something you missed, and more often they have underestimated the verification.

The short version